Ó Tolvan Data 2008
Sirp is a computer program for loudspeaker measurements. It can record the frequency response, both amplitude and phase. It does so by playing logarithmic sine sweeps to the loudspeaker and recording the response with a microphone. From the log sweeps an impulse response is calculated. This impulse response can be truncated so that reflections from the surroundings can be eliminated. For high frequencies this is equivalent to having an anechoic chamber.
Sirp can also measure the distortion of the loudspeaker as a function of frequency.
Measurements from Sirp can be imported into Basta! version 188.8.131.52 and higher.
The signal processing in Sirp is illustrated by the figure below.
A logarithmic sweep is fed to the loudspeaker and recorded by the microphone. The spectrum (aka the fast Fourier transform (FFT)) of the microphone signal is compared to a reference spectrum of the sweep. This reference signal can come directly from the spectrum, or from a loopback signal on the soundcard. The relation between the microphone signal and the reference signal reveals the frequency response of the loudspeaker-microphone chain. This frequency response is transformed back to the time domain by the inverse fast Fourier transform (IFFT), resulting in the impulse response of the loudspeaker-microphone chain. The impulse response can be windowed such that late reflections from the room are discarded. The spectrum of this truncated impulse response then results in the anechoic response of the loudspeaker-microphone chain.
The Sirp main window has four panels. Panel A shows the input signals, panel B the impulse response on a log scale, panel C the truncaded impulse responses of the individual harmonics, and panel D the response and distortion using the truncated impulse response(s).
Panel A shows the raw signals from the soundcard, ie the input signal from the device under test (DUT), and the reference signal, typically a loopback signal from the soundcard output. The signal sent to the soundcard can also be shown here. Select which signals to show from the menu View|waveforms.
Sirp derives the impulse response of the DUT from the signals in panel A. This impulse response is shown in panel B. The y axis is logarithmic and expressed in dB. The cursors are used to select which part of the impulse response that should be used in panels C and D. Typically, the left green cursor is put on top of the main peak of the impulse response. The red and blue cursors defines the length of the excerpt of the signal that is used in the following processing steps.
In panel C1 the excerpt form panel B is shown, but using a linear y-axis. For distortion measurements, Sirp also determines which parts of the impulse response that belong to the harmonic’s impulse responses and show them in panels C2 – C5 (C4 and C5 are not shown here).
In panel D1 the Fourier transform of the impulse responses in panel C1-C5 is shown. If the truncating window in panel B1 is adjusted so that room reflections are discarded, the curves in panel D1 represent the anechoic response of the loudspeaker. Panel D2 shows the harmonics numbers 2-5 relative to harmonic number 1, expressed in percent. Panel D3 shows the phase.
A Sirp measurement of a tweeter on a baffle. The pseudo-anechoic response of the tweeter and distortion is shown in panels D.
There are different approaches on how to adjust the cursors in panel B. The selection of strategy depends on the available acoustic environment, required measurement distance and which frequencies that are of interest. For loudspeaker measurement, there are mainly three different frequency ranges that makes different measurement methods preferable:
The “tweeter range”, ie the range from approximately 500-1000 Hz and upwards can be measured under pseudo-anechoic conditions with Sirp. By finding a location for the measurement that results at least an added pathway of some 1 to 2 meters (~3-6 ms) for the first reflection, all reflections from the room can be removed from the impulse response in panel B. The remaining early parts of the impulse response will be those of the loudspeaker driver and edge diffraction from the box. The microphone should be placed at rather a far distance, as a rule of thumb at least two baffle widths.
The range from 100-300 Hz to 700-1400 Hz is typically the most difficult frequency range to measure. In this range, the baffle step is important, so the microphone will need to be placed at some distance (typically at least two baffle widths), but now the early reflections from the room are required to arrive as late as about 10-20 ms, which corresponds to an added pathway of several meters. This is in some cases not possible. Experimenting with different placements in the room and the settings in panel B can be very rewarding, though. It if moving the loudspeaker-microphone combination in the room causes a peak in the impulse response to move, one can be rather confident in that this peak was a room reflection. Including or excluding it from the impulse response illustrates its effect on the response.
The range from very low frequencies to where the baffle step becomes important at 100-300 Hz if preferably measured in the close field of the loudspeaker driver or vent. Here, the proximity to the driver will make it much louder than the reflections from the room. By adding an appropriate attenuation and delay to the signal, the far-field response can be well estimated. For these measurements, a very long window can be selected. Woofers/vents also typically have a rather long impulse response, so a window length of 100 ms or more is often appropriate.
Sampling rate: Sets the sampling rate of the soundcard. You will probably want to set this rate to the highest sampling rate of the soundcard. On Windows XP and Vista all sounds are resampled to a common rate for all programs, and this resampling procedure is typically not of a sufficiently high quality for measurement purposes. On Windows Vista this sampling rate is visible (and selectable) under Control panel|Sound recording and playback tabs on the “advanced” tabs.
Bandwidth limit: Frequency sweeps in Sirp do not go all the way up to half the sampling rate. By limiting the range slightly, there is a marging for imperfect soundcard anti-alias filters, and it also reduces the frequency response ripple at high frequencies. A limit of approximately 0.8333 and a sampling rate of 48000 Hz set the limit to 20000 Hz.
Record and play settings: Selects the usage of left and right channels. There are also settings for reversing the polarity of the input, if loopback is to be used and if Sirp should try to set zero delay at the impulse response peak. The latter is particularly useful for measurements without a loopback connection.
Recording and playback soundcards: For systems with multiple soundcards, the soundcard can be selected here. It is wise to use the same soundcard for playback and recording.
Smoothing: The frequency response curves can be smoothed to reduce the fluctuations in the response.
Type of measurement: Frequency response or impedance. For impedance measurements, the series resistance should be entered.
Sweep settings: The FFT size determines the duration of the sweep. It roughly occupies half of the playback/recording time and the duration of the entire measurement is the number of samples in the FFT. For example an FFT size of 65536 and a sampling rate of 48000 Hz results in a measurement time of 65536/48000=1.37 seconds.
Silence before recording: Allows you to set a “dead time” before the sweep starts. This might be very useful as it can give you a short moment to hold up a microphone or make a connection after starting the recording.
Sirp allows for three different main types of measurements. The connections for these are illustrated in the figure below.
By using a direct loopback from output A to input B, the delay in the soundcard can be compensated for. This assures that the phase information is accurate, since no delay apart from the loudspeaker/microphone path is added.
Some soundcards have a high-quality built-in microphone amplifier, but the microphone amplifiers of most standard soundcards is typically of rather poor quality. Also, many soundcards only have a mono microphone input, or cannot handle a microphone signal on one input and a line level signal on the other input. Therefore, an external microphone amplifier might be needed, as in the figure.
For simple frequency response measurements, the power amplifier can also sometimes be omitted, depending on the driving capabilities of the soundcard. For high quality measurements, however, the power amplifier can be considered as mandatory.
The loopback can be omitted, but this introduces an unknown delay in the loop, rendering the phase information less useful. This is not the recommended configuration, but it can be used for quick and dirty measurements when phase is not important.
The electrical impedance can be measured using the connection above. This configuration requires a loopback connection and a series resistor. Important note: In this connection, the power amplifier could potentially destroy the input of the soundcard. In some cases it is better to omit the power amplifier and let the output of the soundcard drive the resistor directly. An alternative is to make sure that the inputs of the soundcard are protected by a series resistor, or by using a low-power power amplifier that is not capable of delivering more voltage than the input can take (typically a few volts).
Consider the spectrogram below. In this, the fundamental can be seen as a thick black stripe, whose frequency increases exponentially with time. The harmonic distortion can be seen as weaker (more gray) stripes at 2, 3, 4, 5 etc times the fundamental frequency. The fundamental results in an impulse response after doing the IFFT.
In order to understand the result of the distortion harmonics, we should first consider what would happen with the image if the system has a delay but no distortion. Obviously, the traces fron 2nd, 3rd etc harmonics would not be present, and the fundamental would be shifted sideways, towards a later time. Now, looking again at the distortion components they occur at frequencies 2, 3, 4, 5 etc times higher than the fundamental. But it can be realized that they also can be seen as a copy of the fundamental trace, but occurring earlier in time. This relation is a direct consequence of the exponential shape of the sweep. This in turn leads to that the distortion traces will appear as impulses in the impulse response appearing before the main impulse. So, by cutting out the appropriate parts of the impulse response before the main impulse, and looking at the spectra for these, the amplitude of the each of the harmonics of the distortion can be determined as a function of frequency.
Sirp does not presently offer any means of “stitching” several response curves together, nor can it show curves from multiple measurements for comparison. Basta! ver 184.108.40.206 and higher however, can load multiple curves so that they can be compared.
None so far